def __init__(self, run_data, plots_dir='plots'):
self.plots_dir = plots_dir
# because .sdram_controller_data_width may fail for unimplemented modules
def except_none(func):
try:
return func()
except:
return None
# gather results into tabular data
column_mappings = {
'name':
lambda d: d.config.name,
'sdram_module':
lambda d: d.config.sdram_module,
'sdram_data_width':
lambda d: d.config.sdram_data_width,
'bist_alternating':
lambda d: d.config.bist_alternating,
'num_generators':
lambda d: d.config.num_generators,
'num_checkers':
lambda d: d.config.num_checkers,
'bist_length':
lambda d: getattr(d.config.access_pattern, 'bist_length', None),
'bist_random':
lambda d: getattr(d.config.access_pattern, 'bist_random', None),
'pattern_file':
lambda d: getattr(d.config.access_pattern, 'pattern_file', None),
'length':
lambda d: d.config.length,
'generator_ticks':
lambda d: getattr(d.result, 'generator_ticks', None
), # None means benchmark failure
'checker_errors':
lambda d: getattr(d.result, 'checker_errors', None),
'checker_ticks':
lambda d: getattr(d.result, 'checker_ticks', None),
'ctrl_data_width':
lambda d: except_none(lambda: d.config.sdram_controller_data_width
),
'sdram_memtype':
lambda d: except_none(lambda: d.config.sdram_memtype),
'clk_freq':
lambda d: d.config.sdram_clk_freq,
}
columns = {
name: [mapping(data) for data in run_data]
for name, mapping, in column_mappings.items()
}
self._df = df = pd.DataFrame(columns)
# replace None with NaN
df.fillna(value=np.nan, inplace=True)
# compute other metrics based on ticks and configuration parameters
df['clk_period'] = 1 / df['clk_freq']
# bandwidth is the number of bits per time
# in case with N generators/checkers we actually process N times more data
df['write_bandwidth'] = (8 * df['length'] * df['num_generators']) / (
df['generator_ticks'] * df['clk_period'])
df['read_bandwidth'] = (8 * df['length'] * df['num_checkers']) / (
df['checker_ticks'] * df['clk_period'])
# efficiency calculated as number of write/read commands to number of cycles spent on writing/reading (ticks)
# for multiple generators/checkers multiply by their number
df['cmd_count'] = df['length'] / (df['ctrl_data_width'] / 8)
df['write_efficiency'] = df['cmd_count'] * df['num_generators'] / df[
'generator_ticks']
df['read_efficiency'] = df['cmd_count'] * df['num_checkers'] / df[
'checker_ticks']
df['write_latency'] = df[df['bist_length'] == 1]['generator_ticks']
df['read_latency'] = df[df['bist_length'] == 1]['checker_ticks']
# boolean distinction between latency benchmarks and sequence benchmarks,
# as thier results differ significanly
df['is_latency'] = ~pd.isna(df['write_latency'])
assert (df['is_latency'] == ~pd.isna(df['read_latency'])).all(), \
'write_latency and read_latency should both have a value or both be NaN'
# data formatting for text summary
self.text_formatters = {
'write_bandwidth': bandwidth_fmt,
'read_bandwidth': bandwidth_fmt,
'write_efficiency': efficiency_fmt,
'read_efficiency': efficiency_fmt,
'write_latency': clocks_fmt,
'read_latency': clocks_fmt,
}
# data formatting for plot summary
self.plot_xticks_formatters = {
'write_bandwidth':
FuncFormatter(lambda value, pos: bandwidth_fmt(value)),
'read_bandwidth':
FuncFormatter(lambda value, pos: bandwidth_fmt(value)),
'write_efficiency':
PercentFormatter(1.0),
'read_efficiency':
PercentFormatter(1.0),
'write_latency':
ScalarFormatter(),
'read_latency':
ScalarFormatter(),
}
